Home / Articles / Specialist agents for every team: long-lived agents that keep one skill, design, UX, code, security, behaviour policies or writing, with a person who owns it, so that new teams start from our best work instead of from scratch, and why that makes expertise that never scaled into something you can hire, even from another company
Specialist agents for every team: long-lived agents that keep one skill, design, UX, code, security, behaviour policies or writing, with a person who owns it, so that new teams start from our best work instead of from scratch, and why that makes expertise that never scaled into something you can hire, even from another company
By Dinis Cruz · 2026-11-16 · article v1.0.0 · newsroom v0.14.121 · agentsspecialist-agentsdesignuxsecurityabpexpertisebusiness-modelagenticcellsarticle
Abstract: With the number of agents and sub-teams we now run, I keep seeing the same thing: when I spin up a new team, its developer does not follow our procedures, its designer does not start from our best designs, and its UX repeats mistakes we have already fixed. That makes sense: a new team starts from scratch and has not benefited from all the work before it. So I want a few central teams, made of long-lived sessions, each keeping one skill and the know-how behind it: design, UX, code, security, Agent Behaviour Policies, and writing. They act as consultants to every other team: they bootstrap new teams, review their work, comment on their pull requests, and keep the guidance up to date, and each has a person who owns the skill. Now that our agents can talk to each other through lanes, this works. An agent that has read everything is a compression of what we know, and it is far more efficient to start from it. The writing agent of this newsroom is the example: because it has read everything we have written and keeps the guidance on how to write, it reviewed the book and drafted its introduction. And there is a business model in it. Imagine hiring the specialist agent of a company you respect, with its logic and access to its people behind it, through a lane, and paying per review. That is the old idea of professions that never scaled, now with a team behind them: one person, anywhere, with deep expertise, steering an agent that does the volume.
In short
- New teams start from scratch, so they repeat mistakes: procedures not followed, designs not as good as ours, UX errors we have fixed.
- The fix: a few long-lived specialist agents, design, UX, code, security, behaviour policies and writing, each with a person who owns the skill.
- They act as consultants: they bootstrap new teams, review, comment and keep the guidance current.
- An agent that has read everything is a compression of what we know, and starting from it is far more efficient.
- It is billable: hire another company's specialist agent through a lane, and pay per review.
- Expertise that never scaled now has a team behind it: one person, anywhere, steering an agent that does the volume.
Why new teams start badly
We now run a lot of agents, in many teams: PrompTugal.news with an agent per town, a team per AgentiCCells site, the RiskMandate console, four iOS apps from one workspace (the intersection). And every time I spin up a new team, I see the same thing. The developer does not follow the right procedures. The designer does not produce work as good as it could be, straight away. The UX repeats mistakes we have already fixed elsewhere.
That makes complete sense. A new team starts from scratch. Its agents have not benefited from all the work we have done, the design principles we settled, the release gates we built, the wording rules, the behaviour policies, so they rediscover them, slowly, or not at all.
Specialist agents that keep one skill
So I want to start having central teams, made of specific long-lived sessions, each keeping one skill and the know-how behind it:
- Design: an agent that maintains a design repository for SGit, with our best examples, tokens and patterns, and that has read and captured the good design principles we already have. I can work with that agent directly, make sure it has understood them, and then it carries them across every team.
- UX: what readers and users trip on, and how to avoid it. We still make too many UX mistakes, and better guidance would prevent most of them.
- Code: the procedures, the gates, the release flow and the tests that every repository should follow (the team's operating system).
- Security: reviews, keys, secrets and leak scans, the way our security agents already review batches of commits.
- Agent Behaviour Policies: a policy for every agent and every pipeline, written the same way (CI behaviour policies).
- Writing: voice, structure and wording, the job this newsroom's writing agent already does.
Each one has a person who owns the skill: someone who sets the direction, checks the agent has understood, and decides what changes in the guidance.
Consultants to every other agent
These agents do not do the other teams' work. They act as consultants to it, the way a security or design function supports teams of people today (don't micromanage your agents):
- Bootstrap: a new team gets a pack on day one, what our design looks like, how we release, what a policy must say, instead of starting from nothing.
- Review: any team can ask for a review, and the specialist answers with comments and, where it is allowed, a proposed fix.
- Comment: on pull requests, on drafts, on designs, through the same lanes and repositories every agent already uses.
- Keep the guidance current: when a team finds a better way, the specialist captures it and every other team gets it.
This only works now because our agents can talk to each other, through lanes that keep a record and keys that say who sent what (stop copying and pasting between your agents). A review is a signed message on a lane, and a decision is a page.
A compression of what we know
An agent that has read everything in a domain is a compression of the data and of the information: the principles, the examples, the mistakes and why they were mistakes. It is far more efficient to operate from that compression than to make every team rebuild it.
The writing agent of this newsroom is the example I know best. Because it reads everything we write, and keeps the guidance on how to write, it probably writes the best English of all our agents. When I needed a review of the book, I came to it, and it gave me really good material, and it drafted the introduction. That is what a specialist agent is: not a new model, not training, but an agent with the right context, the right guidance and the right record behind it.
Expertise you can hire, even from another company
This becomes scalable, and billable. Imagine a company whose work I really value. Today, getting their input means a meeting, a contract, weeks. Tomorrow, I could hook my team to their specialist agent: one that has their logic, their guidance and their record, and access to their people when it needs them. My team sends work through a lane; their agent reviews it; their expert signs what needs signing; I pay per review.
Nothing about that needs a new model. It needs the stack we already run: a key for every agent, a lane to send work and receive it, a behaviour policy that says what the specialist may touch, and a record of every review (an inbox only your agents can write to).
Professions that never scaled
This goes back to an idea I keep coming back to: the opportunity is not only in new jobs, but in jobs that never scaled (jobs that never scaled). On-demand expertise was always limited by the expert's hours, and by where they lived. A brilliant designer in a small village, a retired security lead, someone who knows a regulation better than anyone, rarely found the work or the data that needed them.
Now one person, anywhere, can steer a specialist agent that does most of the work, and serve many teams at once: ours, our customers' and other companies'. The expert is paid for judgement; the agent does the volume. That is, fundamentally, what an AgentiCCells mini agentic team is: a small team of agents, powered by a person who knows what good looks like (selling mini agentic teams).
The next phase
This is, I think, the next phase of our work:
- Start with design, code and behaviour policies, where new teams are bootstrapped least well today, then UX, security and writing.
- Give each specialist a repository and a lane: guidance, examples, a bootstrap pack, and an inbox for reviews.
- Make every new team start from the packs, and ask for a review before its first release.
- Measure it: how many review comments a new team gets in its first week, before and after.
- Then offer it outside: a review on the AgentiCCells ladder, and, later, specialist agents other companies can hire.
What this does not claim
- The specialist teams do not exist yet as described; today, the newsroom's writing agent and our security reviewers are the closest examples.
- Hiring another company's agent is an idea, not a product; it needs keys, lanes, policies and a way to pay that work across companies.
- "The best English of all our agents" is my impression, not a measurement.
Where this comes from
My own thinking, 16 November 2026, after spinning up several new teams. Earlier on this site: the intersection, jobs that never scaled, don't micromanage your agents, stop copying and pasting between your agents, the team's operating system, CI behaviour policies, an inbox only your agents can write to, selling mini agentic teams.
Peer review Not reviewed yet. Review it
Threads
Builds on
- The intersection: SGit, SHarness, Agent Behaviour Policies, RiskMandate and AgentiCCells as one stack of separate companies, now with real customers, real usage and real examples, one ladder of offers on every site, and a plan to align them all SGit, SHarness, ABP, RiskMandate and AgentiCCells as one stack of separate companies, with real usage and one ladder of offers on every site.
- The team's operating system: web-summit.sgit.ai, the one site where six people and their agents ran Web Summit, with a brief and a command centre for each person, a key and an inbox for each agent, the board as files, and the outside world on a map web-summit.sgit.ai: one site where six people and their agents ran Web Summit, in four layers built on git.
- CI behaviour policies: to create one repository and one DNS record per request, our site factory's pipeline holds the power to create, change or delete every repository in the organisation and to point any subdomain anywhere, and why pipelines now need the same behaviour policies as agents Our site factory's pipeline needs one repository and one DNS record per request, yet holds power over the whole organisation. Pipelines need ABPs too.
- Don't micromanage your agents: what changed for Vicky Wang when she stopped reading everything her agents did and started managing them, how her team was backed by agents from another team, and why every company will want specialist agents that support all its teams Vicky Wang stopped reading everything her agents did and started managing them, backed by specialist agents from another team: a model for companies.
- Stop copying and pasting between your agents: if you are the wire between your AI sessions you are the bottleneck, and the fix is agents that talk to each other through channels that keep a record, structure that scales your decisions, and outputs you read instead of everything If you copy and paste between AI sessions, you are the wire, and the bottleneck. Let agents talk through recorded channels; read outputs, not everything.
- An inbox only your agents can write to: a conductor publishes one append lane per agent, each lane's token sealed to that agent's own key, so everything needed to talk privately is public and only the right agent can use it A conductor publishes one append lane per agent, each token sealed to that agent's key: a private channel built from public material.
- The jobs that never scaled: on-demand expertise, the work before and after the meeting, and why a historian, a librarian or a consultant can now work for every project that needs one What never scaled is on-demand expertise: the before and after the meeting. Agents can now carry them, for any expert, anywhere.
- Selling mini agentic teams: a business that sets up small agentic teams for companies, sized from one agent to a network, with a control centre to see them and to start and stop them, PrompTugal as the first customer and Agent Behaviour Policies as the most demanding version A business that sets up small agentic teams for companies, with a control centre to start and stop them; PrompTugal first, ABPs on steroids.
Continued by
- The cell is the unit: positioning AgentiCCells around small, focused cells, each with an objective, a deliverable and a behaviour policy, each with a page that sells it, a free recipe and three ways to get help, a catalogue that grows one need at a time, in each industry's own words, and why we should run our own work in cells first The cell is the unit of AgentiCCells: one focused job with an objective, a deliverable and a behaviour policy, each with a page that sells it.
- Our workflow on Odoo 20, built and measured: a bills agent with no Odoo credential, a policy check that allowed two proposals and stopped three, a signed approval, a narrow broker, a grant measured from Odoo itself by a new module, and what Odoo's default user role quietly allows Our workflow on Odoo 20, built in 98 seconds: a keyless agent, a policy check, a signed approval, a narrow broker and a grant measured from Odoo.
- A CRM for us and for our agents: why the next stage is a CRM of our own, built on vaults with an iOS app, why we will build it by using it first for our own team and our own fleet of agents, how the open-source and agentic CRMs compare, and the two teams it needs, one that builds it and one that uses it Our next stage is a CRM of our own, on vaults with an iOS app, for us and our agents. Twenty leads the field; agents propose, people send.
Article No. 432 · start a reading list with it
Reviews
No reviews yet.